Non-vitamin K oral anticoagulants in nonvalvular atrial fibrillation: a network meta-analysis

Abstract

Objectives. We conducted a systematic review and network meta-analysis of randomized controlled trials (RCTs) including the comparison of non-vitamin K antagonist oral anticoagulants (NOACs) with warfarin for patients with atrial fibrillation (AF). Design. Network meta-analysis. Two authors independently extracted data. All authors evaluated overall confidence in the evidence. Results. Eighteen RCTs included in our review, a total of 78,796 patients with AF, with sample sizes from 90 to 21,105 patients. Apixaban 5 mg (OR: 0.79, 95% CI: 0.66 to 0.95), dabigatran 110 mg (0.91, 0.74–1.12), dabigatran 150 mg (0.66, 0.53–0.82), edoxaban 60 mg (0.87, 0.74–1.02), and rivaroxaban 20 mg (0.88, 0.74–1.03) reduced the risk of stroke or systemic embolism compared with warfarin. Dabigatran 150 mg had the highest P-score for reducing stroke or systemic embolic events. The risk of haemorhagic stroke and all-cause mortality was lower with all NOACs than with warfarin. Apixaban 5 mg (0.69, 0.60–0.80), dabigatran 110 mg (0.80, 0.69–0.93), dabigatran 150 mg (0.93, 0.80–1.08), edoxaban 30 mg (0.46, 0.40–0.54), and edoxaban 60 mg (0.78, 0.69–0.90) reduced the risk of major bleeding compared with warfarin. Edoxaban 30 mg had the highest P-score for reducing major bleeding. The plots of P-scores rank showed that apixaban offered the most favorable balance of efficacy and safety. Conclusions. This study adds an attempt for treatment ranking of both efficacy and safety outcomes. Future trials comparing directly NOACs are needed in order to provide conclusive proofs for these results and not only circumstantial evidence offered by a network meta-analysis.

Keywords:

• Non-vitamin K oral anticoagulants
• atrial fibrillation
• network meta-analysis
• Systematic review
• Anticoagulation

Introduction

Over the last decade, several non-vitamin K antagonist oral anticoagulants (NOACs), which selectively inhibit thrombin or activated factor X have been developed for the anticoagulation treatment of patients with atrial fibrillation (AF). The 4 large phase III randomized controlled trials (RCTs) (RE-LY, ROCKET-AF, ARISTOTLE and ENGAGE AF-TIMI 48), as well as a large number of phase II RCTs and several meta-analyses have favored the use of NOACs over vitamin K antagonists in patients with non-valvular AF [1–14].

Evidence-based guidance about the selection of NOACs for AF is not definitive. The comparative effectiveness of treatment regimens that include stepwise approaches is unclear, especially regarding high-risk patient groups, partially because of the scarcity of direct head-to-head trials. Furthermore, the effects of NOACs may have been overestimated as some patients treated with warfarin in clinical trials maintained without the therapeutic INR target of 2.0–3.0 [6,10]. For this purpose, a network meta-analysis (NMA) was performed in order to rank the NOACs and warfarin for each outcome.

Materials and methods

Data searches and study selection

This NMA was registered in PROSPERO with registration number: CRD42017078915 and was reported according to pertinent reporting guidelines for this type of study design [15]. We searched for systematic reviews (SRs) examining the outcomes related to NOACs in AF treatment, when compared to warfarin.

A purposive literature search for SRs and RCTs took place from inception to September 3rd 2017. We screened MEDLINE (via PubMed), Cochrane Library and PROSPERO for SRs and clinicaltrials.gov in order to find further RCTs, only in English language. A more comprehensive search strategy was applied in MEDLINE, using medical subject headings (MeSH) (Supplementary material 1). For all included studies, reference lists were also examined. Two independent authors (CA and ID) screened the titles and abstracts against the eligibility criteria for inclusion.

We searched for SRs and meta-analyses, relevant to our topic and analyzed the full texts to extract the eligible RCTs. A third reviewer (ABH) intervened when there was a disagreement between the other two authors.

Data extraction and quality assessment

The extraction items from the two independent reviewers include: citation details; objectives of the included review; type of review; participant details; setting and context; number of databases sourced and searched; date range of database searching; number of studies, types of studies and country of origin of studies included in each review; instrument used to appraise the primary studies and the rating of their quality; outcomes reported that are relevant; method of synthesis/analysis employed to synthesize the evidence; major conclusions; metric used and effect size (for meta-analyses); confidence intervals (for meta-analyses). Authors were contacted in case of missing data. If the requested data could not be retrieved, the study was not included in the analysis. Data from the individual RCTs were extracted from the included SRs for the NMA.

The risk of bias of the retrieved RCTs, was assessed with the Cochrane risk of bias tool [16]. Six risk of bias domains were assessed, including sequence generation, allocation concealment, blinding of participants and study personnel, blinding of outcome assessments, incomplete outcome data, and selective outcome reporting.

Statistical methods (network Meta-analysis)

NMA was conducted to assess the efficacy and tolerability in two primary (stroke or systemic embolism, major bleeding) and six secondary outcomes (clinically relevant non-major bleeding, gastrointestinal bleeding, myocardial infarction, haemorhagic stroke, ischaemic stroke, all-cause mortality). Direct evidence from two-armed and three-armed RCTs were used to construct treatment networks for each outcome and indirect effect measures were calculated to incorporate the pooled effect of each treatment node. Random-effect NMA was performed based on a graph theoretical approach within a frequentist framework [17]. In the absence of direct comparisons in RCTs, another method to investigate the comparative effect of the interventions of interest is to perform indirect comparisons using a common comparator, warfarin in our case. NMA was performed with R version 3.3.3. using package “netmeta”. Ranking interventions was determined by using P-scores. For each outcome, P-scores allow ranking the treatments on a continuous 0–1 scale and so, it was determined which treatment was the best, the second best, etc [18]. Homogeneity and consistency were assessed by decomposing the Q statistic into variation of the effect estimates within designs (heterogeneity) and between designs (inconsistency). Inconsistency between direct and indirect evidence was investigated by “node-splitting” [19].

Results

Search results

We found eighteen RCTs [1–10,20–27] included a total of 78,796 patients with AF, with sample sizes varying from 90 to 21,105 (Figure 1). None of them was head-to-head trial. Table 1 presents PICOS elements of the individual RCTs. Figure 2 shows the patterns of comparison between the different treatments for two primary outcomes (stroke or systemic embolic events and major bleeding). Both networks have similar geometry, with warfarin acting as the common comparator. Among all the included trials, only 4 (22%) were double blinded, and only 2 (11%) had reported no funding from the industry (Supplementary material 2).

Figure 1. Summary of evidence search and selection.

Figure 2. (A) Network plot for stroke or systemic embolic events. (B) Network plot for major bleeding. The size of nodes is proportional to the number of studies.

Table 1. PICOS Elements of the individual RCTs.

RCTs	Sample size	Population	Interventions in the treatment group	Interventions in the control group	Patients with INR in the therapeutic range	Primary outcome
Ezekowitz 2007 (Petro) [5]	502	Atrial fibrillation	Dabigatran 150 mg × 2	Warfarin (^$INR of 2 to 3)	57.2%	All bleeding events
Connolly 2009 (RE-LY) [1]	18113	Atrial fibrillation	Dabigatran 110 or 150 mg × 2	Warfarin (INR of 2 to 3)	64%	Stroke or systemic embolism
NCT00973323 (2009) [27]	100	Atrial fibrillation	*Rivaroxaban 2.5 or 5 or 10 mg × 1	Warfarin (INR of 2 to 3)	N/A	All bleeding events
NCT00973245 (2009) [26]	102	Atrial fibrillation	*Rivaroxaban 10 or 15 or 20 mg × 1	Warfarin (INR of 2 to 3)	N/A	All bleeding events
NCT01136408 (2010) [25]	166	Atrial fibrillation	Dabigatran 110 or 150 mg × 2	Warfarin (INR of 2 to 3)	N/A	All bleeding events
Weitz 2010 [8]	1146	Atrial fibrillation	Edoxaban 30 or 60 mg × 1	Warfarin (INR of 2 to 3)	64%	μajor plus clinically relevant non-major bleeding events and elevated hepatic enzymes and/or bilirubin
Patel 2011 (ROCKET AF) [3]	14264	Atrial fibrillation	Rivaroxaban 20 mg × 1 or 15 mg in ^#GFR: 30–49	Warfarin (INR of 2 to 3)	58%	Stroke or systemic embolism
Granger 2011 (ARISTOTLE) [4]	18201	Atrial fibrillation	Apixaban 5 mg × 2	Warfarin (INR of 2 to 3)	66%	Stroke or systemic embolism
Ogawa 2011 (ARISTOTLE-J) [7]	222	Atrial fibrillation	Apixaban 2.5 or 5 mg × 2	Warfarin (INR of 2 to 3); 2.0–2.6 if age ≥70 years	60%	Major and non-major clinically relevant bleeding
Chung 2011 [9]	235	Atrial fibrillation	Edoxaban 30 or 60 mg × 1	Warfarin (INR of 2 to 3)	57.1%	All bleeding events
Hori 2012 (J-ROCKET AF) [6]	1278	Atrial fibrillation	*Rivaroxaban 15 mg × 1 or 10 mg in GFR: 30–49	Warfarin (INR of 2 to 3); *1.6–2.6 if age ≥70 years	65%	Safety: Major and non-major clinically relevant bleeding Efficacy: Stroke or systemic embolism
Nin 2013 [24]	90	*Undergoing Ablation	Dabigatran 110 mg × 2	Warfarin (INR of 2 to 3)	N/A	All bleeding events
Yamashita 2012 [10]	536	Atrial fibrillation	Edoxaban 30 or 60 mg × 1	Warfarin (INR of 2 to 3); *1.6–2.6 if age ≥70 years	73%	All bleeding events
Giugliano 2013 (ENGAGE AF-TIMI 48) [2]	21105	Atrial fibrillation	Edoxaban 30 or 60 mg × 1	Warfarin (INR of 2 to 3)	68.4%	Stroke or systemic embolism
Cappato 2014 [21]	1584	*Undergoing elective cardioversion	Rivaroxaban 20 mg × 1 or 15 mg in GFR: 30–49	Warfarin (INR of 2 to 3)	77.6%	τhe composite of stroke, transient ischaemic attack, peripheral embolism, μι, and cardiovascular death
Cappato 2015 [20]	248	*Undergoing Ablation	Rivaroxaban 20 mg × 1	Warfarin (INR of 2 to 3)	79.8%	Major bleeding events
Kuwahara 2016 [23]	200	*Undergoing Ablation	Apixaban 2.5 or 5 mg × 2	Warfarin (INR of 2 to 3)	N/A	Stroke, transient ischemic attack, silent cerebral infarction, or major bleeding
Calkins 2017 [22]	704	*Undergoing Ablation	Dabigatran 150 mg × 2	Warfarin (INR of 2 to 3)	66%	Major bleeding events

Bold indicates RCTs which were included in the network meta-analyses.

*Exclusion criteria of the studies from network meta-analyses.

# GFR: Glomerular filtration rate, ^$INR: International normalized ratio.

Network Meta-analysis

Warfarin was used as common comparator since it was the most common treatment among trials. Comparisons between NOACs and warfarin on the risk of stroke or systemic embolism and major bleeding are shown in Figure 3. Dabigatran at 150mg was found to have the highest P-score for stroke or systemic embolism (followed by apixaban 5mg, edoxaban 60mg, rivaroxaban 20mg, dabigatran 110mg, warfarin and edoxaban 30mg). Different results were observed as regards the risk of major bleeding: Edoxaban, dosed at 30mg, had the highest P-score for major bleeding events. The remaining treatments were ranked as follows: apixaban 5mg, edoxaban 60mg, dabigatran 110mg, dabigatran 150mg, warfarin and rivaroxaban 20mg. Heterogeneity and inconsistency for all outcomes was zero. Odds ratios as calculated from the fitted models for the two primary outcomes are summarized in Table 2.

Figure 3. Comparisons between non-vitamin K antagonists and warfarin on the (A) stroke or systemic embolic events, and (B) major bleeding.

Table 2. Odds ratios with 95% confidence intervals as calculated from the fitted models for major bleeding and stroke or systemic embolism.

	Apixaban 5 mg	Dabigatran 110 mg	Dabigatran 150 mg	Edoxaban 30 mg	Edoxaban 60 mg	Rivaroxaban 20 mg	Warfarin
A. Major bleeding
Apixaban 5 mg	–	0.87 (0.70, 1.07)	0.75 (0.61, 0.92)	1.49 (1.21, 1.84)	0.88 (0.73, 1.08)	0.68 (0.55, 0.83)	0.69 (0.60, 0.80)
Dabigatran 110 mg		–	0.86 (0.74, 1.00)	1.72 (1.39, 2.13)	1.02 (0.83, 1.25)	0.78 (0.63, 0.96)	0.80 (0.69, 0.93)
Dabigatran 150 mg			–	2.00 (1.62, 2.47)	1.19 (0.97, 1.44)	0.91 (0.74, 1.11)	0.93 (0.80, 1.08)
Edoxaban 30 mg				–	0.59 (0.51, 0.70)	0.45 (0.37, 0.56)	0.46 (0.40, 0.54)
Edoxaban 60 mg					–	0.76 (0.63, 0.93)	0.78 (0.69, 0.90)
Rivaroxaban 20 mg						–	1.03 (0.89, 1.19)
Warfarin							–
B. Stroke or systemic embolism
Apixaban 5 mg	–	0.87 (0.66, 1.14)	1.20 (0.90, 1.60)	0.70 (0.55, 0.88)	0.91 (0.72, 1.16)	0.90 (0.70, 1.16)	0.79 (0.66, 0.95)
Dabigatran 110 mg		–	1.38 (1.10, 1.74)	0.80 (0.62, 1.03)	1.05 (0.81, 1.36)	1.04 (0.80, 1.35)	0.91 (0.74, 1.12)
Dabigatran 150 mg			–	0.58 (0.44, 0.76)	0.76 (0.58, 1.00)	0.75 (0.57, 0.99)	0.66 (0.53, 0.82)
Edoxaban 30 mg				–	1.31 (1.12, 1.53)	1.30 (1.04, 1.62)	1.14 (0.98, 1.32)
Edoxaban 60 mg					–	0.99 (0.79, 1.25)	0.87 (0.74, 1.02)
Rivaroxaban 20 mg						–	0.88 (0.74, 1.03)
Warfarin							–

Statistically significant odds ratios are in bold.

All the safety events showed that edoxaban 30mg was the treatment causing less bleeding episodes. Moreover, rankings for stroke events demonstrated that haemorrhagic and ischemic stroke were less frequent under dabigatran 150mg (Supplementary material 3 for all secondary outcomes). The plots of P-scores rank (Figure 4(A)) show that apixaban offered a favourable balance of efficacy and safety. As regards the myocardial infarction, dabigatran at the dose of 150mg was related with a significantly increased risk, while dabigatran 110mg was close to statistical significance. On the other hand, concerning the risk of haemorhagic stroke, all NOACs were better than warfarin.

Figure 4. Plots of P-scores rank for (A) two primary outcomes; (B) gastrointestinal bleeding and clinically relevant non-major bleeding; (C) ischaemic and haemorrhagic stroke.

Discussion

Our NMA of RCTs shows that NOACs appear to be at least as effective as warfarin in reducing the risk of stroke in patients with atrial fibrillation. Apixaban 5 mg and edoxaban 60 mg reduced more the risk of bleeding episodes compared with warfarin, while maintain efficient in reducing the risk of stroke. Although the RCTs have included patient populations with different clinical characteristics and this should not be interpreted as a head-to-head comparison between the NOACs, this work adds an attempt for treatment ranking for both efficacy and safety outcomes to the current literature.

The inclusion of a great number of RCTs does not affect the heterogeneity and the inconsistency, as they were equal to zero for all outcomes. Contrary to other recent meta-analyses [28], our aim was not to include RCTs comparing drugs that are not indicated for the antithrombotic treatment of AF (antiplatelets) [10]. The inclusion of these RCTs in a NMA would inaccurately affect its results. To our knowledge, our NMA has the biggest number of included primary studies compared to the already existing NMA, which had included only the above category of RCTs.

Our findings about the overall efficacy and safety of NOACs are consistent with previously published meta-analyses [29–31]. As regards dabigatran, trials show that different cut-off values of achieved drug levels may play a role in the efficacy and safety of this drug. [32]. This may reduce the convenience of this νOAC and increase its cost, but it can be used for individualization of the dosage in certain cases. Furthermore, there have been doubts about the efficacy data for rivaroxaban from ROCKET-AF [33]. However, the FDA concluded that the effects on stroke or bleeding rates were minimal. The above evidences may not be aggravating for these NOACs. However, the same evidences increase further the need for trials, which will compare NOACs directly, and conclude to which NOAC should be used at each patients’ subgroup in the everyday clinical practice. Consequently, conclusive proofs will be provided and not only circumstantial evidence offered by a network meta-analysis.

Study limitations

The main limitations of this NMA is the number of studies and the complete lack of direct comparisons among the NOACs. All treatments are compared to warfarin or their respective substance at a lower dose. This deprives the constructed networks from crucial information needed to infer on the consistency of the network and the reliability of the results. The current NMA was conducted despite the violation of the transitivity assumption. The fact that all RCTs were funded by industry certainly cause more favorable results due to reporting bias. No meta-regression analysis was attempted to investigate the effect of patient characteristics in individual studies on the pooled outcome.

Conclusions

This study adds an attempt for treatment ranking for both efficacy and safety outcomes. However, no comparative conclusions should be drawn from this NMA. Future trials directly comparing NOACs are needed to infer in efficacy and investigate potential effect modifiers that could be used for patient-targeted therapy.

Acknowledgments

CA: Proposed the structure of the paper, critically appraised the paper and made the final suggestions

ID: Proposed the idea, proposed the statistical analysis and formulate the paper

EA: Proposed the statistical analysis

FIE: Critically appraised the paper, made the final suggestions

PV: Critically appraised the paper, made the final suggestions

ABH: Supervised the statistical analysis, critically appraised the paper and made the final suggestions

VK: Critically appraised the paper, made the final suggestions.

Disclosure statement

No potential conflict of interest was reported by the authors.